Prototype SIMD extending pairwise add instructions (#3466)

As proposed in https://github.com/WebAssembly/simd/pull/380, using the opcodes
used in LLVM and V8. Since these opcodes overlap with the opcodes of
i64x2.all_true and i64x2.any_true, which have long since been removed from the
SIMD proposal, this PR also removes those instructions.

17 files changed, 102 insertions, 77 deletions

diff --git a/src/binaryen-c.cpp b/src/binaryen-c.cpp
index b6e14d694..59ca3aebc 100644
--- a/src/binaryen-c.cpp
+++ b/src/binaryen-c.cpp
@@ -580,8 +580,6 @@ BinaryenOp BinaryenDotSVecI16x8ToVecI32x4(void) {
   return DotSVecI16x8ToVecI32x4;
 }
 BinaryenOp BinaryenNegVecI64x2(void) { return NegVecI64x2; }
-BinaryenOp BinaryenAnyTrueVecI64x2(void) { return AnyTrueVecI64x2; }
-BinaryenOp BinaryenAllTrueVecI64x2(void) { return AllTrueVecI64x2; }
 BinaryenOp BinaryenShlVecI64x2(void) { return ShlVecI64x2; }
 BinaryenOp BinaryenShrSVecI64x2(void) { return ShrSVecI64x2; }
 BinaryenOp BinaryenShrUVecI64x2(void) { return ShrUVecI64x2; }
diff --git a/src/binaryen-c.h b/src/binaryen-c.h
index 33ada13b1..a32d9ff1c 100644
--- a/src/binaryen-c.h
+++ b/src/binaryen-c.h
@@ -486,8 +486,6 @@ BINARYEN_API BinaryenOp BinaryenMaxSVecI32x4(void);
 BINARYEN_API BinaryenOp BinaryenMaxUVecI32x4(void);
 BINARYEN_API BinaryenOp BinaryenDotSVecI16x8ToVecI32x4(void);
 BINARYEN_API BinaryenOp BinaryenNegVecI64x2(void);
-BINARYEN_API BinaryenOp BinaryenAnyTrueVecI64x2(void);
-BINARYEN_API BinaryenOp BinaryenAllTrueVecI64x2(void);
 BINARYEN_API BinaryenOp BinaryenShlVecI64x2(void);
 BINARYEN_API BinaryenOp BinaryenShrSVecI64x2(void);
 BINARYEN_API BinaryenOp BinaryenShrUVecI64x2(void);
@@ -528,6 +526,7 @@ BINARYEN_API BinaryenOp BinaryenCeilVecF64x2(void);
 BINARYEN_API BinaryenOp BinaryenFloorVecF64x2(void);
 BINARYEN_API BinaryenOp BinaryenTruncVecF64x2(void);
 BINARYEN_API BinaryenOp BinaryenNearestVecF64x2(void);
+// TODO: Add extending pairwise adds to C and JS APIs once merged
 BINARYEN_API BinaryenOp BinaryenTruncSatSVecF32x4ToVecI32x4(void);
 BINARYEN_API BinaryenOp BinaryenTruncSatUVecF32x4ToVecI32x4(void);
 BINARYEN_API BinaryenOp BinaryenTruncSatSVecF64x2ToVecI64x2(void);
diff --git a/src/gen-s-parser.inc b/src/gen-s-parser.inc
index d4e1afe38..c564f3974 100644
--- a/src/gen-s-parser.inc
+++ b/src/gen-s-parser.inc
@@ -4,8 +4,8 @@
 
 #ifdef INSTRUCTION_PARSER
 #undef INSTRUCTION_PARSER
-char op[28] = {'\0'};
-strncpy(op, s[0]->c_str(), 27);
+char op[30] = {'\0'};
+strncpy(op, s[0]->c_str(), 29);
 switch (op[0]) {
   case 'a': {
     switch (op[1]) {
@@ -800,6 +800,17 @@ switch (op[0]) {
                 goto parse_error;
               case 'x': {
                 switch (op[9]) {
+                  case 'a': {
+                    switch (op[28]) {
+                      case 's':
+                        if (strcmp(op, "i16x8.extadd_pairwise_i8x16_s") == 0) { return makeUnary(s, UnaryOp::ExtAddPairwiseSVecI8x16ToI16x8); }
+                        goto parse_error;
+                      case 'u':
+                        if (strcmp(op, "i16x8.extadd_pairwise_i8x16_u") == 0) { return makeUnary(s, UnaryOp::ExtAddPairwiseUVecI8x16ToI16x8); }
+                        goto parse_error;
+                      default: goto parse_error;
+                    }
+                  }
                   case 'm': {
                     switch (op[13]) {
                       case 'h': {
@@ -1554,6 +1565,17 @@ switch (op[0]) {
                         goto parse_error;
                       case 'x': {
                         switch (op[9]) {
+                          case 'a': {
+                            switch (op[28]) {
+                              case 's':
+                                if (strcmp(op, "i32x4.extadd_pairwise_i16x8_s") == 0) { return makeUnary(s, UnaryOp::ExtAddPairwiseSVecI16x8ToI32x4); }
+                                goto parse_error;
+                              case 'u':
+                                if (strcmp(op, "i32x4.extadd_pairwise_i16x8_u") == 0) { return makeUnary(s, UnaryOp::ExtAddPairwiseUVecI16x8ToI32x4); }
+                                goto parse_error;
+                              default: goto parse_error;
+                            }
+                          }
                           case 'm': {
                             switch (op[13]) {
                               case 'h': {
@@ -2298,20 +2320,9 @@ switch (op[0]) {
           }
           case 'x': {
             switch (op[6]) {
-              case 'a': {
-                switch (op[7]) {
-                  case 'd':
-                    if (strcmp(op, "i64x2.add") == 0) { return makeBinary(s, BinaryOp::AddVecI64x2); }
-                    goto parse_error;
-                  case 'l':
-                    if (strcmp(op, "i64x2.all_true") == 0) { return makeUnary(s, UnaryOp::AllTrueVecI64x2); }
-                    goto parse_error;
-                  case 'n':
-                    if (strcmp(op, "i64x2.any_true") == 0) { return makeUnary(s, UnaryOp::AnyTrueVecI64x2); }
-                    goto parse_error;
-                  default: goto parse_error;
-                }
-              }
+              case 'a':
+                if (strcmp(op, "i64x2.add") == 0) { return makeBinary(s, BinaryOp::AddVecI64x2); }
+                goto parse_error;
               case 'b':
                 if (strcmp(op, "i64x2.bitmask") == 0) { return makeUnary(s, UnaryOp::BitmaskVecI64x2); }
                 goto parse_error;
diff --git a/src/ir/cost.h b/src/ir/cost.h
index c218cbaab..ef0a2bde1 100644
--- a/src/ir/cost.h
+++ b/src/ir/cost.h
@@ -189,8 +189,6 @@ struct CostAnalyzer : public OverriddenVisitor<CostAnalyzer, Index> {
       case AllTrueVecI32x4:
       case BitmaskVecI32x4:
       case NegVecI64x2:
-      case AnyTrueVecI64x2:
-      case AllTrueVecI64x2:
       case BitmaskVecI64x2:
       case AbsVecF32x4:
       case NegVecF32x4:
@@ -206,6 +204,10 @@ struct CostAnalyzer : public OverriddenVisitor<CostAnalyzer, Index> {
       case FloorVecF64x2:
       case TruncVecF64x2:
       case NearestVecF64x2:
+      case ExtAddPairwiseSVecI8x16ToI16x8:
+      case ExtAddPairwiseUVecI8x16ToI16x8:
+      case ExtAddPairwiseSVecI16x8ToI32x4:
+      case ExtAddPairwiseUVecI16x8ToI32x4:
       case TruncSatSVecF32x4ToVecI32x4:
       case TruncSatUVecF32x4ToVecI32x4:
       case TruncSatSVecF64x2ToVecI64x2:
diff --git a/src/ir/features.h b/src/ir/features.h
index 6dff96398..aa87bd5bd 100644
--- a/src/ir/features.h
+++ b/src/ir/features.h
@@ -57,8 +57,6 @@ inline FeatureSet get(UnaryOp op) {
     case AnyTrueVecI32x4:
     case AllTrueVecI32x4:
     case NegVecI64x2:
-    case AnyTrueVecI64x2:
-    case AllTrueVecI64x2:
     case AbsVecF32x4:
     case NegVecF32x4:
     case SqrtVecF32x4:
diff --git a/src/js/binaryen.js-post.js b/src/js/binaryen.js-post.js
index 128e148e2..6cdf26e32 100644
--- a/src/js/binaryen.js-post.js
+++ b/src/js/binaryen.js-post.js
@@ -416,8 +416,6 @@ function initializeConstants() {
     'MaxSVecI32x4',
     'MaxUVecI32x4',
     'NegVecI64x2',
-    'AnyTrueVecI64x2',
-    'AllTrueVecI64x2',
     'ShlVecI64x2',
     'ShrSVecI64x2',
     'ShrUVecI64x2',
@@ -1828,12 +1826,6 @@ function wrapModule(module, self = {}) {
     'neg'(value) {
       return Module['_BinaryenUnary'](module, Module['NegVecI64x2'], value);
     },
-    'any_true'(value) {
-      return Module['_BinaryenUnary'](module, Module['AnyTrueVecI64x2'], value);
-    },
-    'all_true'(value) {
-      return Module['_BinaryenUnary'](module, Module['AllTrueVecI64x2'], value);
-    },
     'shl'(vec, shift) {
       return Module['_BinaryenSIMDShift'](module, Module['ShlVecI64x2'], vec, shift);
     },
diff --git a/src/literal.h b/src/literal.h
index bf8937533..5b590e07f 100644
--- a/src/literal.h
+++ b/src/literal.h
@@ -576,8 +576,6 @@ public:
   Literal extMulLowUI32x4(const Literal& other) const;
   Literal extMulHighUI32x4(const Literal& other) const;
   Literal negI64x2() const;
-  Literal anyTrueI64x2() const;
-  Literal allTrueI64x2() const;
   Literal shlI64x2(const Literal& other) const;
   Literal shrSI64x2(const Literal& other) const;
   Literal shrUI64x2(const Literal& other) const;
diff --git a/src/passes/Print.cpp b/src/passes/Print.cpp
index b921fb074..f9d469faf 100644
--- a/src/passes/Print.cpp
+++ b/src/passes/Print.cpp
@@ -957,12 +957,6 @@ struct PrintExpressionContents
       case NegVecI64x2:
         o << "i64x2.neg";
         break;
-      case AnyTrueVecI64x2:
-        o << "i64x2.any_true";
-        break;
-      case AllTrueVecI64x2:
-        o << "i64x2.all_true";
-        break;
       case BitmaskVecI64x2:
         o << "i64x2.bitmask";
         break;
@@ -1008,6 +1002,18 @@ struct PrintExpressionContents
       case NearestVecF64x2:
         o << "f64x2.nearest";
         break;
+      case ExtAddPairwiseSVecI8x16ToI16x8:
+        o << "i16x8.extadd_pairwise_i8x16_s";
+        break;
+      case ExtAddPairwiseUVecI8x16ToI16x8:
+        o << "i16x8.extadd_pairwise_i8x16_u";
+        break;
+      case ExtAddPairwiseSVecI16x8ToI32x4:
+        o << "i32x4.extadd_pairwise_i16x8_s";
+        break;
+      case ExtAddPairwiseUVecI16x8ToI32x4:
+        o << "i32x4.extadd_pairwise_i16x8_u";
+        break;
       case TruncSatSVecF32x4ToVecI32x4:
         o << "i32x4.trunc_sat_f32x4_s";
         break;
diff --git a/src/tools/fuzzing.h b/src/tools/fuzzing.h
index 7d9131767..09e83031d 100644
--- a/src/tools/fuzzing.h
+++ b/src/tools/fuzzing.h
@@ -2168,9 +2168,7 @@ private:
                                     AnyTrueVecI16x8,
                                     AllTrueVecI16x8,
                                     AnyTrueVecI32x4,
-                                    AllTrueVecI32x4,
-                                    AnyTrueVecI64x2,
-                                    AllTrueVecI64x2),
+                                    AllTrueVecI32x4),
                                make(Type::v128)});
           }
           case Type::funcref:
diff --git a/src/wasm-binary.h b/src/wasm-binary.h
index 382bfbd4f..b2c922dc2 100644
--- a/src/wasm-binary.h
+++ b/src/wasm-binary.h
@@ -897,8 +897,6 @@ enum ASTNodes {
   I64x2WidenLowUI32x4 = 0xc9,
   I64x2WidenHighUI32x4 = 0xca,
   I64x2Neg = 0xc1,
-  I64x2AnyTrue = 0xc2,
-  I64x2AllTrue = 0xc3,
   I64x2Shl = 0xcb,
   I64x2ShrS = 0xcc,
   I64x2ShrU = 0xcd,
@@ -939,6 +937,11 @@ enum ASTNodes {
   F64x2PMin = 0xf6,
   F64x2PMax = 0xf7,
 
+  I16x8ExtAddPairWiseSI8x16 = 0xc2,
+  I16x8ExtAddPairWiseUI8x16 = 0xc3,
+  I32x4ExtAddPairWiseSI16x8 = 0xa5,
+  I32x4ExtAddPairWiseUI16x8 = 0xa6,
+
   I32x4TruncSatSF32x4 = 0xf8,
   I32x4TruncSatUF32x4 = 0xf9,
   F32x4ConvertSI32x4 = 0xfa,
diff --git a/src/wasm-interpreter.h b/src/wasm-interpreter.h
index 019a56db1..3d963269d 100644
--- a/src/wasm-interpreter.h
+++ b/src/wasm-interpreter.h
@@ -488,10 +488,6 @@ public:
         return value.bitmaskI32x4();
       case NegVecI64x2:
         return value.negI64x2();
-      case AnyTrueVecI64x2:
-        return value.anyTrueI64x2();
-      case AllTrueVecI64x2:
-        return value.allTrueI64x2();
       case BitmaskVecI64x2:
         WASM_UNREACHABLE("unimp");
       case AbsVecF32x4:
@@ -522,6 +518,14 @@ public:
         return value.truncF64x2();
       case NearestVecF64x2:
         return value.nearestF64x2();
+      case ExtAddPairwiseSVecI8x16ToI16x8:
+        WASM_UNREACHABLE("unimp");
+      case ExtAddPairwiseUVecI8x16ToI16x8:
+        WASM_UNREACHABLE("unimp");
+      case ExtAddPairwiseSVecI16x8ToI32x4:
+        WASM_UNREACHABLE("unimp");
+      case ExtAddPairwiseUVecI16x8ToI32x4:
+        WASM_UNREACHABLE("unimp");
       case TruncSatSVecF32x4ToVecI32x4:
         return value.truncSatToSI32x4();
       case TruncSatUVecF32x4ToVecI32x4:
diff --git a/src/wasm.h b/src/wasm.h
index 1d5642625..12aacd2ed 100644
--- a/src/wasm.h
+++ b/src/wasm.h
@@ -174,8 +174,6 @@ enum UnaryOp {
   AllTrueVecI32x4,
   BitmaskVecI32x4,
   NegVecI64x2,
-  AnyTrueVecI64x2,
-  AllTrueVecI64x2,
   BitmaskVecI64x2,
   AbsVecF32x4,
   NegVecF32x4,
@@ -191,6 +189,10 @@ enum UnaryOp {
   FloorVecF64x2,
   TruncVecF64x2,
   NearestVecF64x2,
+  ExtAddPairwiseSVecI8x16ToI16x8,
+  ExtAddPairwiseUVecI8x16ToI16x8,
+  ExtAddPairwiseSVecI16x8ToI32x4,
+  ExtAddPairwiseUVecI16x8ToI32x4,
 
   // SIMD conversions
   TruncSatSVecF32x4ToVecI32x4,
diff --git a/src/wasm/literal.cpp b/src/wasm/literal.cpp
index f526cebf3..49d3cbf63 100644
--- a/src/wasm/literal.cpp
+++ b/src/wasm/literal.cpp
@@ -1868,12 +1868,6 @@ Literal Literal::allTrueI32x4() const {
 Literal Literal::bitmaskI32x4() const {
   return bitmask<4, &Literal::getLanesI32x4>(*this);
 }
-Literal Literal::anyTrueI64x2() const {
-  return any_true<2, &Literal::getLanesI64x2>(*this);
-}
-Literal Literal::allTrueI64x2() const {
-  return all_true<2, &Literal::getLanesI64x2>(*this);
-}
 
 template<int Lanes,
          LaneArray<Lanes> (Literal::*IntoLanes)() const,
diff --git a/src/wasm/wasm-binary.cpp b/src/wasm/wasm-binary.cpp
index 70aaa6a61..e616493af 100644
--- a/src/wasm/wasm-binary.cpp
+++ b/src/wasm/wasm-binary.cpp
@@ -4836,14 +4836,6 @@ bool WasmBinaryBuilder::maybeVisitSIMDUnary(Expression*& out, uint32_t code) {
       curr = allocator.alloc<Unary>();
       curr->op = NegVecI64x2;
       break;
-    case BinaryConsts::I64x2AnyTrue:
-      curr = allocator.alloc<Unary>();
-      curr->op = AnyTrueVecI64x2;
-      break;
-    case BinaryConsts::I64x2AllTrue:
-      curr = allocator.alloc<Unary>();
-      curr->op = AllTrueVecI64x2;
-      break;
     case BinaryConsts::I64x2Bitmask:
       curr = allocator.alloc<Unary>();
       curr->op = BitmaskVecI64x2;
@@ -4904,6 +4896,22 @@ bool WasmBinaryBuilder::maybeVisitSIMDUnary(Expression*& out, uint32_t code) {
       curr = allocator.alloc<Unary>();
       curr->op = NearestVecF64x2;
       break;
+    case BinaryConsts::I16x8ExtAddPairWiseSI8x16:
+      curr = allocator.alloc<Unary>();
+      curr->op = ExtAddPairwiseSVecI8x16ToI16x8;
+      break;
+    case BinaryConsts::I16x8ExtAddPairWiseUI8x16:
+      curr = allocator.alloc<Unary>();
+      curr->op = ExtAddPairwiseUVecI8x16ToI16x8;
+      break;
+    case BinaryConsts::I32x4ExtAddPairWiseSI16x8:
+      curr = allocator.alloc<Unary>();
+      curr->op = ExtAddPairwiseSVecI16x8ToI32x4;
+      break;
+    case BinaryConsts::I32x4ExtAddPairWiseUI16x8:
+      curr = allocator.alloc<Unary>();
+      curr->op = ExtAddPairwiseUVecI16x8ToI32x4;
+      break;
     case BinaryConsts::I32x4TruncSatSF32x4:
       curr = allocator.alloc<Unary>();
       curr->op = TruncSatSVecF32x4ToVecI32x4;
diff --git a/src/wasm/wasm-stack.cpp b/src/wasm/wasm-stack.cpp
index 6af7279bb..776928b88 100644
--- a/src/wasm/wasm-stack.cpp
+++ b/src/wasm/wasm-stack.cpp
@@ -1028,14 +1028,6 @@ void BinaryInstWriter::visitUnary(Unary* curr) {
     case NegVecI64x2:
       o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::I64x2Neg);
       break;
-    case AnyTrueVecI64x2:
-      o << int8_t(BinaryConsts::SIMDPrefix)
-        << U32LEB(BinaryConsts::I64x2AnyTrue);
-      break;
-    case AllTrueVecI64x2:
-      o << int8_t(BinaryConsts::SIMDPrefix)
-        << U32LEB(BinaryConsts::I64x2AllTrue);
-      break;
     case BitmaskVecI64x2:
       o << int8_t(BinaryConsts::SIMDPrefix)
         << U32LEB(BinaryConsts::I64x2Bitmask);
@@ -1084,6 +1076,22 @@ void BinaryInstWriter::visitUnary(Unary* curr) {
       o << int8_t(BinaryConsts::SIMDPrefix)
         << U32LEB(BinaryConsts::F64x2Nearest);
       break;
+    case ExtAddPairwiseSVecI8x16ToI16x8:
+      o << int8_t(BinaryConsts::SIMDPrefix)
+        << U32LEB(BinaryConsts::I16x8ExtAddPairWiseSI8x16);
+      break;
+    case ExtAddPairwiseUVecI8x16ToI16x8:
+      o << int8_t(BinaryConsts::SIMDPrefix)
+        << U32LEB(BinaryConsts::I16x8ExtAddPairWiseUI8x16);
+      break;
+    case ExtAddPairwiseSVecI16x8ToI32x4:
+      o << int8_t(BinaryConsts::SIMDPrefix)
+        << U32LEB(BinaryConsts::I32x4ExtAddPairWiseSI16x8);
+      break;
+    case ExtAddPairwiseUVecI16x8ToI32x4:
+      o << int8_t(BinaryConsts::SIMDPrefix)
+        << U32LEB(BinaryConsts::I32x4ExtAddPairWiseUI16x8);
+      break;
     case TruncSatSVecF32x4ToVecI32x4:
       o << int8_t(BinaryConsts::SIMDPrefix)
         << U32LEB(BinaryConsts::I32x4TruncSatSF32x4);
diff --git a/src/wasm/wasm-validator.cpp b/src/wasm/wasm-validator.cpp
index 745ce8c06..a45be1fd3 100644
--- a/src/wasm/wasm-validator.cpp
+++ b/src/wasm/wasm-validator.cpp
@@ -1866,6 +1866,10 @@ void FunctionValidator::visitUnary(Unary* curr) {
     case FloorVecF64x2:
     case TruncVecF64x2:
     case NearestVecF64x2:
+    case ExtAddPairwiseSVecI8x16ToI16x8:
+    case ExtAddPairwiseUVecI8x16ToI16x8:
+    case ExtAddPairwiseSVecI16x8ToI32x4:
+    case ExtAddPairwiseUVecI16x8ToI32x4:
     case TruncSatSVecF32x4ToVecI32x4:
     case TruncSatUVecF32x4ToVecI32x4:
     case TruncSatSVecF64x2ToVecI64x2:
@@ -1893,11 +1897,9 @@ void FunctionValidator::visitUnary(Unary* curr) {
     case AnyTrueVecI8x16:
     case AnyTrueVecI16x8:
     case AnyTrueVecI32x4:
-    case AnyTrueVecI64x2:
     case AllTrueVecI8x16:
     case AllTrueVecI16x8:
     case AllTrueVecI32x4:
-    case AllTrueVecI64x2:
     case BitmaskVecI8x16:
     case BitmaskVecI16x8:
     case BitmaskVecI32x4:
diff --git a/src/wasm/wasm.cpp b/src/wasm/wasm.cpp
index 436eedb2b..264f2bf83 100644
--- a/src/wasm/wasm.cpp
+++ b/src/wasm/wasm.cpp
@@ -783,6 +783,10 @@ void Unary::finalize() {
     case FloorVecF64x2:
     case TruncVecF64x2:
     case NearestVecF64x2:
+    case ExtAddPairwiseSVecI8x16ToI16x8:
+    case ExtAddPairwiseUVecI8x16ToI16x8:
+    case ExtAddPairwiseSVecI16x8ToI32x4:
+    case ExtAddPairwiseUVecI16x8ToI32x4:
     case TruncSatSVecF32x4ToVecI32x4:
     case TruncSatUVecF32x4ToVecI32x4:
     case TruncSatSVecF64x2ToVecI64x2:
@@ -808,11 +812,9 @@ void Unary::finalize() {
     case AnyTrueVecI8x16:
     case AnyTrueVecI16x8:
     case AnyTrueVecI32x4:
-    case AnyTrueVecI64x2:
     case AllTrueVecI8x16:
     case AllTrueVecI16x8:
     case AllTrueVecI32x4:
-    case AllTrueVecI64x2:
     case BitmaskVecI8x16:
     case BitmaskVecI16x8:
     case BitmaskVecI32x4: